530087 - improve xptcinvoke for linux on arm

Status: RESOLVED DUPLICATE of bug 532198

(Core :: XPCOM, defect)

Description

[Brad Lassey [:blassey] (use needinfo?)]

We got a pretty good perf boost in bug 529348 for windows ce and windows mobile, I'd like to see if a similar approach would help maemo.

Comment 1

[Brad Lassey [:blassey] (use needinfo?)]

Attachment: patch

with this patch, I see a 1.68x improvement according to the speed test in TextXPTCInvoke

without patch:
 direct took 9.05 seconds
 invoke took 111.08 seconds
 So, invoke overhead was ~ 102.03 seconds (~ 92%)

with patch:
 direct took 9.12 seconds
 invoke took 69.88 seconds
 So, invoke overhead was ~ 60.76 seconds (~ 87%)

Comment 2

[Brad Lassey [:blassey] (use needinfo?)]

Ts impact:

without: 10613, 11228, 10777, 11224 => 10960.5 average 
with: 10130, 10252, 11130, 10867 => 10602.25 average

so, by my testing its a 358.25ms win

Comment 3

[Mike Beltzner [:beltzner, not reading bugmail]]

a=beltzner to land this on trunk once it passes review; find me again (I'll be around!) for a192 if that works out cleanly.

Comment 4

[Justin Dolske [:Dolske]]

FWIW, I've done a bit of browsing on a WinCE trunk nightly with this change, and it seems to work fine.

Comment 5

[Brad Lassey [:blassey] (use needinfo?)]

pushed http://hg.mozilla.org/mozilla-central/rev/0573fb2293c0

Comment 6

[Brad Lassey [:blassey] (use needinfo?)]

backed out

Comment 7

[Brad Lassey [:blassey] (use needinfo?)]

backed out http://hg.mozilla.org/mozilla-central/rev/daa5346f19d6

Comment 8

[Reed Loden [:reed]]

Bug 532198 includes a patch that does a small bit of what the patch in this bug does.

Comment 9

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Comment on attachment 413887 [details] [diff] [review]
patch

Unapproving due to bounce

Comment 10

[Mark Finkle (:mfinkle) (use needinfo?)]

Jacob - any thoughts on when you will be working on this bug? I'm curious as to when we might see it for a fennec release.

Comment 11

[Jacob Bramley [:jbramley]]

Yep, I'm looking at it now but it keeps being interrupted by unrelated things, and it's being slow due to native ARM build times. (That isn't really a problem for Trace Monkey builds, but it is for XPCOM.)

I have a couple of patches that just need testing, though I don't know why you saw that failure before and I suspect whatever caused it will be in my patch too.

Comment 12

[Stuart Parmenter]

can we close this/dup it to 532198?